The server's 24 tools cost ~21,411 tokens (tiktoken cl100k) at tools/list — the heaviest of 20 popular servers we measured (https://github.com/lCrazyblindl/lap/blob/main/docs/MCP-LEADERBOARD.md). The descriptions are terse (nice!); the weight is schemas: API-update-page-markdown ~1469 tokens, API-post-search ~1258 — the same Notion object subtrees inlined into tool after tool.
Measured: hoisting identical repeated subtrees (≥120 chars, seen ≥2×) into shared $defs saves ~14,811 tokens (~69%) on its own, before any semantic slimming. The 2026-07-28 MCP spec explicitly allows $ref/$defs in inputSchema (SEP-2106), so the compatibility risk that used to argue for inlining is going away. Separately, 12 tools declare parameters without descriptions. Method + script: https://github.com/lCrazyblindl/lap/blob/main/docs/UPSTREAM-ISSUES.md.
Happy to contribute the $defs hoisting or the measurement harness. (Disclosure: I maintain lap, the measurement tool; MIT, no product.)
The server's 24 tools cost ~21,411 tokens (tiktoken cl100k) at
tools/list— the heaviest of 20 popular servers we measured (https://github.com/lCrazyblindl/lap/blob/main/docs/MCP-LEADERBOARD.md). The descriptions are terse (nice!); the weight is schemas:API-update-page-markdown~1469 tokens,API-post-search~1258 — the same Notion object subtrees inlined into tool after tool.Measured: hoisting identical repeated subtrees (≥120 chars, seen ≥2×) into shared
$defssaves ~14,811 tokens (~69%) on its own, before any semantic slimming. The 2026-07-28 MCP spec explicitly allows$ref/$defsininputSchema(SEP-2106), so the compatibility risk that used to argue for inlining is going away. Separately, 12 tools declare parameters without descriptions. Method + script: https://github.com/lCrazyblindl/lap/blob/main/docs/UPSTREAM-ISSUES.md.Happy to contribute the $defs hoisting or the measurement harness. (Disclosure: I maintain
lap, the measurement tool; MIT, no product.)